feat(bench): pin ferrflow to single thread + report parallel stat with runner cores

[BryanFRD]

Closes #144. Depends on FerrLabs/FerrFlow#597 (the --jobs flag).

What

Since FerrLabs/FerrFlow#476 made per-package planning parallel, the competitive benchmark's ferrflow numbers scaled with the runner's core count — non-reproducible and unfair against single-threaded competitors.

• The competitive comparison now runs ferrflow with --jobs 1, so its numbers are single-threaded, reproducible, and apples-to-apples with semantic-release / changesets / standard-version / commit-and-tag-version.
• A second, all-cores pass runs per fixture/command and is recorded separately under ferrflow_parallel (kept out of the benchmarks object and the comparison table), with a top-level runner_cores so the real-world speedup is visible without distorting the fair comparison.

latest.json additions

{
  "runner_cores": 4,
  "benchmarks": { /* ferrflow pinned to --jobs 1, unchanged shape */ },
  "ferrflow_parallel": {
    "mono-large-release-dry": { "median_ms": 72.0, "stddev_ms": 1.2 }
  }
}

Parallel raw files live in raw/parallel/ so they don't leak into the existing benchmarks glob or the markdown table. Additive only — compare.sh / format-release.sh and their bats tests are unaffected (bash -n clean). The ferrlabs.com perf page rendering of ferrflow_parallel + runner_cores follows in a FerrFlow-Cloud PR.